[openstack-dev] [nova] Action items from the "Road to live upgrade" summit session
Sean Dague
sdague at linux.vnet.ibm.com
Tue Oct 30 21:05:55 UTC 2012
On 10/30/2012 12:12 PM, Dan Smith wrote:
> Hi all,
>
> I have gone through the etherpad[*] for the "Road to live upgrade"
> summit session and pulled out what I believe are the necessary action
> items. Below are the artifacts/blueprints (with names) I plan to create
> as a result, but I wanted to float them here for comments before I do.
>
> - Add a wiki page to document best practices for writing new database
> migrations. Some of these are documented in the "Suggestions" portion
> of the etherpad, but this will require some work by people that
> really understand the pitfalls and perils of that work.
While looking for the right place to add this, I came across this
existing wiki page (very sparse) -
http://wiki.openstack.org/ReviewChecklist. Seems like the right place we
should start? I think there were a lot of concrete comments on what a
good migration looks like as part of that which we should write down.
> - Add a nova blueprint for inserting sample data into the database
> prior to migration testing (migration-testing-with-data). The group
> identified the need to do migration tests with more than an empty
> database to catch consistency issues.
>
> - Add a QA blueprint for multinode testing (multinode-testing).
>
> - Add a QA blueprint for doing nightly runs of Grenade (nightly-upgrade-
> testing).
>
> - Add a nova blueprint for versioning the RPC wireline format (version-
> rpc-messages). The group confirmed the findings of the previous
> session on trusted RPC with respect to the need for versioned RPC
> wireline messages so that future improvements can be made without
> requiring cluster-atomic upgrades.
>
> - Add a nova blueprint for controlling the version used for outbound
> RPC messages (rpc-version-control). Right now, a version N+1 node
> can receive and handle version N messages, but can only send version
> N+1 messages. Allowing an admin to lock down the RPC version until
> all nodes have been upgraded allows for graceful cluster-wide
> upgrades.
I'm not sure we fully got to the solution here, but we at least know
that the fact that RPC versioning is only a receiver function isn't good
enough, as sometimes nova-compute needs to cross talk.
Whether this is an admin thing, or a negotiated thing is I think
something that can be sorted out once someone finally starts tackling
the problem.
> - Add a QA (?) blueprint for mixed-version RPC testing (mixed-
> rpc-version-testing). This was a proposed attempt to specifically
> test a commit against the previous to ensure that the RPC layers can
> communicate (i.e. pass all the regular RPC tests). I figure this will
> be a scary script to splice things together, and that it could/would
> be run as part of gate or nightly testing.
>
> - Add a nova blueprint for versioning notifications (versioned-
> notifications). The group identified the need to version notification
> messages for the same reason as RPC messages.
>
> Comments?
>
> *: https://etherpad.openstack.org/grizzly-live-upgrade
-Sean
--
Sean Dague
IBM Linux Technology Center
email: sdague at linux.vnet.ibm.com
alt-email: sldague at us.ibm.com
More information about the OpenStack-dev
mailing list